Enterprise AI at the Decision Point: A U.S.-Focused Systematic Review of Productivity, ROI, Workforce Change, and Governance

Enterprise AI assistants and agentic systems have expanded rapidly since 2022, prompting organizations to make increasingly consequential decisions about technology investment, enterprise-wide deployment, productivity expectations, and workforce design. Two claims now dominate much of the public and managerial discussion: that enterprise AI will generate substantial and durable productivity and financial returns, and that it will enable organizations to operate effectively with fewer human workers. Both claims increasingly influence organizational decision making, yet neither has been established consistently across real-world evidence. This systematic review therefore examines what enterprise AI has measurably delivered—particularly in the United States—and whether the available evidence is sufficiently complete to support high-stakes enterprise-governance decisions. ISACA’s CGEIT domains are used as an established governance lens for evaluating decision readiness, particularly for investment, production deployment, scale-up, budget expansion, benefits realization, risk oversight, and workforce restructuring. Following PRISMA 2020, we reviewed evidence published from January 2022 through 16 September 2026. The search identified 471 records, of which 61 studies were included after screening. The evidence base comprised peer-reviewed and field experiments, U.S. government sources, official institutional evaluations, high quality analyst research, and clearly labeled vendor materials. Complementary administrative workforce data and targeted searches were used to examine actual AI expenditure, realized ROI, workforce substitution, and the adequacy of evidence available for CGEIT-aligned governance decisions. The findings challenge both dominant claims. U.S. Census data show AI use in 18% of firms (32% employment-weighted). Among AIusing firms, 66% reported augmentation-only use, while 2% reported AI-related employment decreases. Because the latter is self-reported firm attribution, it should not be interpreted as the true rate of AI-driven workforce substitution or as a complete estimate of AI-associated workforce reduction. Causal studies demonstrated meaningful but highly task-dependent productivity effects, including approximately 15% more customer-service issues resolved per hour and 26.08% more completed coding tasks, but also null effects and performance losses in other settings. Thus, observed productivity gains do not generalize uniformly across tasks, tools, or organizational contexts. More importantly for investment decisions, independently measured financial ROI for named enterprise AI assistants was not identified. Among 350 large organizations already using agents, approximately 80% reported some workforce reduction; however, organizations reporting greater workforce reduction did not consistently report higher ROI. In other words, using AI with fewer human workers did not, by itself, demonstrate stronger financial returns. A targeted substitution analysis identified no complete case (n = 0) demonstrating the full evidentiary chain from actual AI cost and budget trade-off to verified workforce replacement and realized ROI. Taken together, the evidence does not support treating either large-scale human-labor substitution or reliably positive enterprise ROI as an established outcome of enterprise AI adoption. The more defensible interpretation is one of uneven augmentation, context-dependent productivity gains, incomplete benefits realization, and major evidence gaps in actual cost, realized ROI, replacement capability, reliability, workforce attribution, and residual risk. These gaps directly constrain CGEIT-aligned executive decisions on whether to move from pilot to production, expand licenses or budgets, scale enterprise deployment, or redesign the workforce around AI. To address this decision problem, the study proposes a minimum evidence package and an evidence-gated sequence for first investment, pilot, production, scale-up, budget expansion, and workforce restructuring.